Kadai Khata
Built at Meta Global AI Developer Hackathon · Oct 26, 2026 · Remote

Kadai Khata is an AI-assisted udhaar (credit) ledger for Tamil Nadu kirana shop owners, who today track customer credit in paper notebooks. The owner types or speaks a Tanglish sentence (for example "Ravi-ku aanju nooru rooba udhaar kuduthen") or photographs a handwritten notebook page, and Meta models turn it into structured ledger drafts. The AI never saves anything: every draft appears on a review slip with deterministic checks (overpayment, new customer, unusual amount, ambiguous name match), and only the owner's Confirm writes to the ledger. Features: Tanglish text parsing with tool calling (several entries per sentence), voice input with an owner-reviewed transcript, notebook-photo digitising with a per-column total check against the totals written on the page, payment reminders in Tamil, Tanglish or English whose amounts are verified against the database, a read-only balance question, and a collections agent that proposes who to remind today. The agent can only read the ledger; a deterministic verifier recomputes every number it claims, and each run is stored in a hash-chained trace that can be replayed offline. It is a mobile-first web app with an English/Tamil toggle, a live demo on fictional data, 161 automated tests (147 offline, 14 browser), and honest LIVE / MOCK / FALLBACK labels on every output. Limitations (small evaluation sets, voice script drift, ephemeral demo disk, no authentication) are documented in the README.